Security

Security
architecture.

How clawd.run protects your data, isolates your agent, and earns your trust. Built on open-source. Auditable by design.

clawd.run is built on OpenClaw, an open-source agent framework. You should know exactly what runs on your behalf, where your data lives, and what guarantees you actually have.

Encryption

Data protection at every layer

In Transit

All connections use TLS 1.2+. API calls to AI providers, Stripe, and internal services are encrypted end-to-end. WebSocket connections to sandboxes use WSS.

At Rest

All stored data is encrypted at rest using AES-256. Sandbox volumes use encrypted storage. Payment data is handled by a PCI DSS Level 1 certified processor.

Agent Credentials

Channel credentials use AES-256-GCM, keyed per-user. API keys are stored as HMAC hashes. Never plaintext, never logged, never sent anywhere unintended.

Isolation

Sandbox architecture

Invite-enabled: Dedicated containers

  • ✓Each user gets their own isolated container
  • ✓Separate filesystem, process space, and network namespace
  • ✓No shared memory or processes between users
  • ✓Sandboxes stopped when inactive, started on demand
  • ✓Agent tools confined to the sandbox

Base access: Shared infrastructure

  • ◌Free users share API gateway infrastructure (no dedicated sandbox)
  • ◌Conversations isolated per user via database security rules
  • ◌No filesystem or tool access. Chat only.
  • ◌Rate limited to prevent abuse (50 messages per day)

Data access

Controls and your rights

Database security rules

Every collection has explicit security rules. Users can only read and write their own data. Admin-only collections are inaccessible from the client.

  • ◆User profile, config, API keys: read/write own data only
  • ◆Conversations: read own, writes via server
  • ◆Usage tracking: read own, write server-side only
  • ◆Admin logs, system config: no client access

Your rights

  • ✓Export: Download all your data from Settings
  • ✓Delete: Clear history or delete your entire account
  • ✓Inspect: Bot config files visible in your dashboard
  • ✓Portable: Your agent config is standard markdown. Transfer to any OpenClaw instance.

AI providers

What they see

When you send a message, it is forwarded to an AI provider. They receive:

  • ◆Your message and recent conversation context
  • ◆System prompt (your agent's SOUL.md and configuration)
  • ◆Tool results if your agent executes tools (invite-enabled)

Most major providers (Anthropic, OpenAI) do not train on API inputs by default. If you use BYOK, data is sent under your own API agreement.

Open source

Audit it yourself

clawd.run is built on OpenClaw. The framework's security model covers sandbox modes, tool policies, authentication, and threat model.

Security docs →Source code →

Found a vulnerability?

Report security issues to security@clawd.run. Do not open a public GitHub issue. We respond within 48 hours.